diff --git a/Android.bp b/Android.bp
index 8a596cd..6937ab5 100644
--- a/Android.bp
+++ b/Android.bp
@@ -110,7 +110,6 @@
         "src/core/CL/kernels/CLDilateKernel.cpp",
         "src/core/CL/kernels/CLDirectConvolutionLayerKernel.cpp",
         "src/core/CL/kernels/CLElementWiseUnaryLayerKernel.cpp",
-        "src/core/CL/kernels/CLElementwiseOperationKernel.cpp",
         "src/core/CL/kernels/CLErodeKernel.cpp",
         "src/core/CL/kernels/CLFFTDigitReverseKernel.cpp",
         "src/core/CL/kernels/CLFFTRadixStageKernel.cpp",
@@ -441,6 +440,7 @@
         "src/core/gpu/cl/kernels/ClActivationKernel.cpp",
         "src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp",
         "src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp",
+        "src/core/gpu/cl/kernels/ClElementwiseKernel.cpp",
         "src/core/gpu/cl/kernels/ClFloorKernel.cpp",
         "src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp",
@@ -795,6 +795,7 @@
         "src/runtime/cpu/operators/CpuReshape.cpp",
         "src/runtime/cpu/operators/CpuSub.cpp",
         "src/runtime/gpu/cl/operators/ClActivation.cpp",
+        "src/runtime/gpu/cl/operators/ClAdd.cpp",
         "src/runtime/gpu/cl/operators/ClConcatenate.cpp",
         "src/runtime/gpu/cl/operators/ClFloor.cpp",
         "utils/CommonGraphOptions.cpp",
diff --git a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
index 55c5fb3..4dd4912 100644
--- a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
+++ b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -35,74 +35,7 @@
 
 namespace experimental
 {
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for addition
- *
- * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
- * @note The function performs an arithmetic addition between two tensors.
- */
-class CLArithmeticAddition : public ICLOperator
-{
-public:
-    /** Default Constructor */
-    CLArithmeticAddition();
-    /** Initialise the kernel's inputs, output and conversion policy.
-     *
-     * Valid configurations (Input1,Input2) -> Output :
-     *
-     *   - (U8,U8)           -> U8
-     *   - (U8,U8)           -> S16
-     *   - (S16,U8)          -> S16
-     *   - (U8,S16)          -> S16
-     *   - (S16,S16)         -> S16
-     *   - (S32,S32)         -> S32
-     *   - (F16,F16)         -> F16
-     *   - (F32,F32)         -> F32
-     *   - (QASYMM8,QASYMM8) -> QASYMM8
-     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
-     *   - (QSYMM16,QSYMM16) -> QSYMM16
-     *
-     * @param[in]      compile_context The compile context to be used.
-     * @param[in, out] input1          First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     *                                 The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
-     * @param[in, out] input2          Second tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     *                                 The input tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
-     * @param[out]     output          Output tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     * @param[in]      policy          Policy to use to handle overflow.
-     * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
-     */
-    void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy,
-                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for addition
-     *
-     * Valid configurations (Input1,Input2) -> Output :
-     *
-     *   - (U8,U8)           -> U8
-     *   - (U8,U8)           -> S16
-     *   - (S16,U8)          -> S16
-     *   - (U8,S16)          -> S16
-     *   - (S16,S16)         -> S16
-     *   - (S32,S32)         -> S32
-     *   - (F16,F16)         -> F16
-     *   - (F32,F32)         -> F32
-     *   - (QASYMM8,QASYMM8) -> QASYMM8
-     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
-     *   - (QSYMM16,QSYMM16) -> QSYMM16
-     *
-     * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     * @param[in] input2   Second tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     * @param[in] output   Output tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
-     * @param[in] policy   Policy to use to handle overflow.
-     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-    // Inherited methods overridden:
-    void run(ITensorPack &tensors) override;
-};
-
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for subtraction
+/** Basic function to run @ref arm_compute::opencl::kernels::ClSaturatedArithmeticKernel for subtraction
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/S32/F16/F32.
  * @note The function performs an arithmetic subtraction between two tensors.
@@ -139,7 +72,7 @@
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy,
                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for subtraction
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClSaturatedArithmeticKernel for subtraction
      *
      * Valid configurations (Input1,Input2) -> Output :
      *
@@ -169,7 +102,7 @@
     void run(ITensorPack &tensors) override;
 };
 
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for division
+/** Basic function to run @ref arm_compute::opencl::kernels::ClSaturatedArithmeticKernel for division
  *
  * @note The tensor data type for the inputs must be F16/F32.
  * @note The function performs an arithmetic division between two tensors.
@@ -205,7 +138,7 @@
     void run(ITensorPack &tensors) override;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for max
+/** Basic function to run @ref arm_compute::opencl::kernels::ClArithmeticKernel for max
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32.
  * @note The function performs a max operation between two tensors.
@@ -226,7 +159,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for max
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClArithmeticKernel for max
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -241,7 +174,7 @@
     void run(ITensorPack &tensors) override;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for min
+/** Basic function to run @ref arm_compute::opencl::kernels::ClArithmeticKernel for min
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32.
  * @note The function performs a max operation between two tensors.
@@ -262,7 +195,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for min
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClArithmeticKernel for min
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -277,7 +210,7 @@
     void run(ITensorPack &tensors) override;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for squared difference
+/** Basic function to run @ref arm_compute::opencl::kernels::ClArithmeticKernel for squared difference
  *
  * @note The tensor data type for the inputs must be QASYMM8/U8/S16/QSYMM16/F16/F32.
  * @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2
@@ -298,7 +231,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for squared difference
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClArithmeticKernel for squared difference
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -313,7 +246,7 @@
     void run(ITensorPack &tensors) override;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for power
+/** Basic function to run @ref arm_compute::opencl::kernels::ClArithmeticKernel for power
  *
  * @note The tensor data type for the inputs must be F16/F32.
  * @note The function performs an elementwise power of in1 to in2 (i.e., out[i] = in1[i] ^ in2[i])
@@ -334,7 +267,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for power
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClArithmeticKernel for power
      *
      * @param[in] input1   First tensor input info. Data types supported: F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: F16/F32.
@@ -350,7 +283,7 @@
 };
 } // namespace experimental
 
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for addition
+/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for addition
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
  * @note The function performs an arithmetic addition between two tensors.
@@ -422,7 +355,7 @@
      */
     void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ConvertPolicy policy,
                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for addition
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClSaturatedArithmeticKernel for addition
      *
      * Valid configurations (Input1,Input2) -> Output :
      *
@@ -456,7 +389,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for subtraction
+/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for subtraction
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/S32/F16/F32.
  * @note The function performs an arithmetic subtraction between two tensors.
@@ -528,7 +461,7 @@
      */
     void configure(const CLCompileContext &compile_context, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ConvertPolicy policy,
                    const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel for subtraction
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClSaturatedArithmeticKernel for subtraction
      *
      * Valid configurations (Input1,Input2) -> Output :
      *
@@ -562,7 +495,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLSaturatedArithmeticOperationKernel for division
+/** Basic function to run @ref opencl::kernels::ClSaturatedArithmeticKernel for division
  *
  * @note The tensor data type for the inputs must be F16/F32.
  * @note The function performs an arithmetic division between two tensors.
@@ -622,7 +555,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for max
+/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for max
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32.
  * @note The function performs a max operation between two tensors.
@@ -663,7 +596,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for max
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for max
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -682,7 +615,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for min
+/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for min
  *
  * @note The tensor data type for the inputs must be U8/QASYMM8/S16/QSYMM16/S32/U32/F16/F32.
  * @note The function performs a max operation between two tensors.
@@ -723,7 +656,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for min
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for min
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/U32/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -742,7 +675,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for squared difference
+/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for squared difference
  *
  * @note The tensor data type for the inputs must be QASYMM8/U8/S16/QSYMM16/F16/F32.
  * @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2
@@ -783,7 +716,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for squared difference
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for squared difference
      *
      * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: same as @p input1.
@@ -802,7 +735,7 @@
     std::unique_ptr<Impl> _impl;
 };
 
-/** Basic function to run @ref CLArithmeticOperationKernel for power
+/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for power
  *
  * @note The tensor data type for the inputs must be F16/F32.
  * @note The function performs an elementwise power of in1 to in2 (i.e., out[i] = in1[i] ^ in2[i])
@@ -843,7 +776,7 @@
      * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
      */
     void configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel for power
+    /** Static function to check if given info will lead to a valid configuration of @ref opencl::kernels::ClArithmeticKernel for power
      *
      * @param[in] input1   First tensor input info. Data types supported: F16/F32.
      * @param[in] input2   Second tensor input info. Data types supported: F16/F32.
diff --git a/arm_compute/runtime/CL/functions/CLLogicalAnd.h b/arm_compute/runtime/CL/functions/CLLogicalAnd.h
index f5d834c..f7038ee 100644
--- a/arm_compute/runtime/CL/functions/CLLogicalAnd.h
+++ b/arm_compute/runtime/CL/functions/CLLogicalAnd.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -51,7 +51,7 @@
      * @param[out]     output          Output tensor. Data types supported: same as @p input1.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLLogicalBinaryKernel
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClLogicalBinaryKernel
      *
      * @param[in] input1 First tensor input info. Data types supported: U8.
      * @param[in] input2 Second tensor input info. Data types supported: same as @p input1.
@@ -65,7 +65,7 @@
 };
 } // namespace experimental
 
-/** Basic function to run @ref CLLogicalBinaryKernel.
+/** Basic function to run @ref arm_compute::opencl::kernels::ClLogicalBinaryKernel.
  *
  * @note The tensor data type for the inputs must be U8.
  * @note The function performs a logical AND operation using the two input tensors.
diff --git a/arm_compute/runtime/CL/functions/CLLogicalOr.h b/arm_compute/runtime/CL/functions/CLLogicalOr.h
index daf7d85..948baee 100644
--- a/arm_compute/runtime/CL/functions/CLLogicalOr.h
+++ b/arm_compute/runtime/CL/functions/CLLogicalOr.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -51,7 +51,7 @@
      * @param[out]     output          Output tensor. Data types supported: same as @p input1.
      */
     void configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLLogicalBinaryKernel
+    /** Static function to check if given info will lead to a valid configuration of @ref arm_compute::opencl::kernels::ClLogicalBinaryKernel
      *
      * @param[in] input1 First tensor input info. Data types supported: U8.
      * @param[in] input2 Second tensor input info. Data types supported: same as @p input1.
@@ -65,7 +65,7 @@
 };
 } // namespace experimental
 
-/** Basic function to run @ref CLLogicalBinaryKernel.
+/** Basic function to run @ref arm_compute::opencl::kernels::ClLogicalBinaryKernel.
  *
  * @note The tensor data type for the inputs must be U8.
  * @note The function performs a logical OR operation using the two input tensors.
diff --git a/arm_compute/runtime/CL/functions/CLPReluLayer.h b/arm_compute/runtime/CL/functions/CLPReluLayer.h
index ab32bcc..1751fda 100644
--- a/arm_compute/runtime/CL/functions/CLPReluLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPReluLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -35,7 +35,7 @@
 
 namespace experimental
 {
-/** Basic function to run @ref CLArithmeticOperationKernel for PRELU
+/** Basic function to run @ref arm_compute::opencl::kernels::ClArithmeticKernel for PRELU
  *
  * @note The function implements an activation layer with the PRELU activation function.
  */
@@ -69,7 +69,7 @@
 };
 } // namespace experimental
 
-/** Basic function to run @ref CLArithmeticOperationKernel for PRELU
+/** Basic function to run @ref opencl::kernels::ClArithmeticKernel for PRELU
  *
  * @note The function implements an activation layer with the PRELU activation function.
  */
diff --git a/docs/00_introduction.dox b/docs/00_introduction.dox
index 3c89cfd..d7ef3dc 100644
--- a/docs/00_introduction.dox
+++ b/docs/00_introduction.dox
@@ -166,7 +166,7 @@
    - @ref NEGEMMLowpMatrixBReductionKernel
  - Removed padding from OpenCL kernels:
    - CLBatchConcatenateLayerKernel
-   - @ref CLElementwiseOperationKernel
+   - CLElementwiseOperationKernel
    - @ref CLBatchNormalizationLayerKernel
    - @ref CLPoolingLayerKernel
    - @ref CLWinogradInputTransformKernel
diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index 76bc527..916cf45 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -57,7 +57,6 @@
 #include "src/core/CL/kernels/CLDilateKernel.h"
 #include "src/core/CL/kernels/CLDirectConvolutionLayerKernel.h"
 #include "src/core/CL/kernels/CLElementWiseUnaryLayerKernel.h"
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
 #include "src/core/CL/kernels/CLErodeKernel.h"
 #include "src/core/CL/kernels/CLFFTDigitReverseKernel.h"
 #include "src/core/CL/kernels/CLFFTRadixStageKernel.h"
diff --git a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp b/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
deleted file mode 100644
index 7c1d940..0000000
--- a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
+++ /dev/null
@@ -1,516 +0,0 @@
-/*
- * Copyright (c) 2018-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/ICLTensor.h"
-#include "src/core/CL/CLValidate.h"
-#include "src/core/common/Validate.h"
-#include "src/core/helpers/AutoConfiguration.h"
-#include "src/core/helpers/WindowHelpers.h"
-#include "support/Cast.h"
-#include "support/StringSupport.h"
-#include <map>
-
-namespace arm_compute
-{
-namespace
-{
-constexpr unsigned int vector_size_byte_opencl = 16;
-
-std::map<ArithmeticOperation, std::string> supported_arithmetic_ops =
-{
-    { ArithmeticOperation::ADD, "ADD" },
-    { ArithmeticOperation::SUB, "SUB" },
-    { ArithmeticOperation::DIV, "DIV" },
-    { ArithmeticOperation::SQUARED_DIFF, "SQUARED_DIFF" },
-    { ArithmeticOperation::MIN, "MIN" },
-    { ArithmeticOperation::MAX, "MAX" },
-    { ArithmeticOperation::POWER, "POWER" },
-    { ArithmeticOperation::PRELU, "PRELU" },
-};
-
-std::map<ArithmeticOperation, std::string> supported_sat_arithmetic_ops =
-{
-    { ArithmeticOperation::ADD, "ADD" },
-    { ArithmeticOperation::SUB, "SUB" },
-};
-
-std::string generate_id_for_tuning_common(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
-{
-    std::string config_id;
-    // Set config_id for enabling LWS tuning
-    config_id = kernel_name;
-    config_id += "_";
-    config_id += lower_string(string_from_data_type(input1.data_type()));
-    config_id += "_";
-    config_id += support::cpp11::to_string(output.dimension(0));
-    config_id += "_";
-    config_id += support::cpp11::to_string(output.dimension(1));
-    return config_id;
-}
-
-Status validate_arguments_with_float_only_supported_rules(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(&input1, &input2, &output);
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input1);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &input2);
-
-    const TensorShape out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
-
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
-
-    // Validate in case of configured output
-    if(output.total_size() > 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output, 1, DataType::F16, DataType::F32);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, output.tensor_shape(), 0),
-                                        "Wrong shape for output");
-    }
-
-    return Status{};
-}
-
-Status validate_arguments_with_arithmetic_rules(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input1);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
-                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
-                                                         DataType::S32, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input2);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input2, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
-                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
-                                                         DataType::S32, DataType::F32);
-
-    const bool is_quantized = is_data_type_quantized(input1.data_type()) || is_data_type_quantized(input2.data_type());
-    if(is_quantized)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &input2);
-
-        if(is_data_type_quantized_symmetric(input1.data_type()))
-        {
-            const int32_t in1_offset = input1.quantization_info().uniform().offset;
-            const int32_t in2_offset = input2.quantization_info().uniform().offset;
-            ARM_COMPUTE_RETURN_ERROR_ON_MSG(in1_offset != 0, "For quantized symmetric, offset must be zero");
-            ARM_COMPUTE_RETURN_ERROR_ON_MSG(in2_offset != 0, "For quantized symmetric, offset must be zero");
-        }
-    }
-
-    const TensorShape out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
-
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
-
-    // Validate in case of configured output
-    if(output.total_size() > 0)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&output);
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
-                                                             DataType::S16, DataType::QSYMM16, DataType::F16,
-                                                             DataType::S32, DataType::F32);
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG((output.data_type() == DataType::U8) && ((input1.data_type() != DataType::U8) || (input2.data_type() != DataType::U8)),
-                                        "Output can only be U8 if both inputs are U8");
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, output.tensor_shape(), 0),
-                                        "Wrong shape for output");
-
-        if(is_quantized)
-        {
-            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &output);
-
-            if(is_data_type_quantized_symmetric(output.data_type()))
-            {
-                const int32_t offset = output.quantization_info().uniform().offset;
-                ARM_COMPUTE_RETURN_ERROR_ON_MSG(offset != 0, "For quantized symmetric, offset must be zero");
-            }
-        }
-    }
-    return Status{};
-}
-
-CLBuildOptions generate_build_options_with_arithmetic_rules(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output, const std::string &operation_string)
-{
-    CLBuildOptions build_opts;
-
-    const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / output.element_size(), output.dimension(0));
-
-    build_opts.add_option("-DDATA_TYPE_IN1=" + get_cl_type_from_data_type(input1.data_type()));
-    build_opts.add_option("-DDATA_TYPE_IN2=" + get_cl_type_from_data_type(input2.data_type()));
-    build_opts.add_option("-DDATA_TYPE_OUT=" + get_cl_type_from_data_type(output.data_type()));
-    build_opts.add_option("-DVEC_SIZE_IN1=" + support::cpp11::to_string(input1.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
-    build_opts.add_option("-DVEC_SIZE_IN2=" + support::cpp11::to_string(input2.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
-    build_opts.add_option("-DVEC_SIZE_OUT=" + support::cpp11::to_string(num_elems_processed_per_iteration));
-    build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(output.dimension(0) % num_elems_processed_per_iteration));
-    build_opts.add_option("-DOP=" + operation_string);
-    if(is_data_type_quantized(input1.data_type()))
-    {
-        const UniformQuantizationInfo iq1info = input1.quantization_info().uniform();
-        const UniformQuantizationInfo iq2info = input2.quantization_info().uniform();
-        const UniformQuantizationInfo oqinfo  = output.quantization_info().uniform();
-
-        build_opts.add_option("-DOFFSET_IN1=" + support::cpp11::to_string(iq1info.offset));
-        build_opts.add_option("-DOFFSET_IN2=" + support::cpp11::to_string(iq2info.offset));
-        build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(oqinfo.offset));
-        build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1info.scale));
-        build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2info.scale));
-        build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oqinfo.scale));
-    }
-    return build_opts;
-}
-
-std::pair<Status, Window> configure_window_arithmetic_common(ITensorInfo &output)
-{
-    const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / output.element_size(), output.dimension(0));
-    Window             win                               = calculate_max_window(output, Steps(num_elems_processed_per_iteration));
-    return std::make_pair(Status{}, win);
-}
-
-std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
-    const TensorShape &out_shape = broadcast_pair.first;
-
-    set_shape_if_empty(output, out_shape);
-
-    if(input1.data_type() == DataType::S16 || input2.data_type() == DataType::S16)
-    {
-        set_format_if_unknown(output, Format::S16);
-    }
-    else if(input1.data_type() == DataType::F16 || input2.data_type() == DataType::F16)
-    {
-        set_format_if_unknown(output, Format::F16);
-    }
-    else if(input1.data_type() == DataType::F32 || input2.data_type() == DataType::F32)
-    {
-        set_format_if_unknown(output, Format::F32);
-    }
-    else if(input1.data_type() == DataType::QASYMM8 || input2.data_type() == DataType::QASYMM8)
-    {
-        set_data_type_if_unknown(output, DataType::QASYMM8);
-    }
-    else if(input1.data_type() == DataType::QASYMM8_SIGNED || input2.data_type() == DataType::QASYMM8_SIGNED)
-    {
-        set_data_type_if_unknown(output, DataType::QASYMM8_SIGNED);
-    }
-    else if(input1.data_type() == DataType::QSYMM16 || input2.data_type() == DataType::QSYMM16)
-    {
-        set_data_type_if_unknown(output, DataType::QSYMM16);
-    }
-
-    return configure_window_arithmetic_common(output);
-}
-
-std::pair<Status, Window> validate_and_configure_window_for_logical_binary_operators(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
-    const TensorShape &out_shape = broadcast_pair.first;
-
-    set_shape_if_empty(output, out_shape);
-    set_data_type_if_unknown(output, DataType::U8);
-
-    // The arithmetic utility functions can be share
-    return configure_window_arithmetic_common(output);
-}
-
-std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
-    const TensorShape &out_shape = broadcast_pair.first;
-    auto_init_if_empty(output, out_shape, 1, input1.data_type());
-    return configure_window_arithmetic_common(output);
-}
-} // namespace
-
-CLElementwiseOperationKernel::CLElementwiseOperationKernel()
-    : _act_info(), _input1(nullptr), _input2(nullptr), _output(nullptr)
-{
-}
-
-void CLElementwiseOperationKernel::configure_common(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
-{
-    configure_common(CLKernelLibrary::get().get_compile_context(), input1, input2, output);
-}
-
-void CLElementwiseOperationKernel::configure_common(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
-{
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(*input1, *input2, *output);
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-
-    _input1 = input1;
-    _input2 = input2;
-    _output = output;
-
-    std::string kernel_name = "elementwise_operation_" + name();
-    if(is_data_type_quantized(input1->data_type()))
-    {
-        kernel_name += "_quantized";
-    }
-
-    // Set kernel build options
-    CLBuildOptions build_opts = generate_build_options(*input1, *input2, *output);
-    if(_act_info.enabled())
-    {
-        build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(_act_info.activation())));
-        build_opts.add_option("-DA_VAL=" + float_to_string_with_full_precision(_act_info.a()));
-        build_opts.add_option("-DB_VAL=" + float_to_string_with_full_precision(_act_info.b()));
-    }
-
-    // Create kernel
-    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
-
-    ICLKernel::configure_internal(win_config.second);
-
-    _config_id = generate_id_for_tuning(kernel_name, *input1, *output);
-}
-
-void CLElementwiseOperationKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
-{
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
-
-    const auto src_0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
-    const auto src_1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
-    auto       dst   = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
-
-    const TensorShape &in_shape1 = src_0->info()->tensor_shape();
-    const TensorShape &in_shape2 = src_1->info()->tensor_shape();
-    const TensorShape &out_shape = dst->info()->tensor_shape();
-
-    bool       can_collapse = true;
-    const bool is_vector    = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
-    if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
-    {
-        can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
-        for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
-        {
-            can_collapse = (in_shape1[d] == in_shape2[d]);
-        }
-    }
-
-    bool   has_collapsed = false;
-    Window collapsed     = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
-
-    const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
-    const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
-
-    Window slice        = collapsed.first_slice_window_3D();
-    Window slice_input1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed);
-    Window slice_input2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed);
-    do
-    {
-        unsigned int idx = 0;
-        add_3D_tensor_argument(idx, src_0, slice_input1);
-        add_3D_tensor_argument(idx, src_1, slice_input2);
-        add_3D_tensor_argument(idx, dst, slice);
-
-        enqueue(queue, *this, slice, lws_hint());
-        ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input1));
-        ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input2));
-    }
-    while(collapsed.slide_window_slice_3D(slice));
-}
-
-/** Logical binary */
-
-void CLLogicalBinaryKernel::configure(const CLCompileContext &compile_context, kernels::LogicalOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_ERROR_THROW_ON(CLLogicalBinaryKernel::validate(op, input1, input2, output));
-    _op = op;
-    configure_common(compile_context, input1, input2, output);
-}
-
-Status CLLogicalBinaryKernel::validate(kernels::LogicalOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
-{
-    ARM_COMPUTE_UNUSED(op);
-    ARM_COMPUTE_ASSERT(op != kernels::LogicalOperation::Unknown && op != kernels::LogicalOperation::Not);
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
-
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input1, 1, DataType::U8);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input1, input2);
-
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_logical_binary_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
-
-    return Status{};
-}
-
-std::string CLLogicalBinaryKernel::name()
-{
-    switch(_op)
-    {
-        case kernels::LogicalOperation::And:
-            return "AND";
-        case kernels::LogicalOperation::Or:
-            return "OR";
-        case kernels::LogicalOperation::Not:
-        /* fall through */
-        default:
-            ARM_COMPUTE_ASSERT(true);
-    }
-    return "";
-}
-
-std::pair<Status, Window> CLLogicalBinaryKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    return validate_and_configure_window_for_logical_binary_operators(input1, input2, output);
-}
-
-CLBuildOptions CLLogicalBinaryKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
-{
-    // The arithmetic utility functions can be share
-    return generate_build_options_with_arithmetic_rules(input1, input2, output, name());
-}
-
-std::string CLLogicalBinaryKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
-{
-    return generate_id_for_tuning_common(kernel_name, input1, output);
-}
-
-/** Arithmetic operations with saturation*/
-
-void CLSaturatedArithmeticOperationKernel::configure(ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy,
-                                                     const ActivationLayerInfo &act_info)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), op, input1, input2, output, policy, act_info);
-}
-
-void CLSaturatedArithmeticOperationKernel::configure(const CLCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output,
-                                                     const ConvertPolicy       &policy,
-                                                     const ActivationLayerInfo &act_info)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_ERROR_THROW_ON(CLSaturatedArithmeticOperationKernel::validate(op, input1, input2, output, policy, act_info));
-    auto padding_info = get_padding_info({ input1, input2, output });
-
-    _policy   = policy;
-    _op       = op;
-    _act_info = act_info;
-    configure_common(compile_context, input1, input2, output);
-    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
-}
-
-Status CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy,
-                                                      const ActivationLayerInfo &act_info)
-{
-    ARM_COMPUTE_UNUSED(op, policy);
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
-    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled() && !is_data_type_float(output->data_type()));
-
-    return Status{};
-}
-
-std::pair<Status, Window> CLSaturatedArithmeticOperationKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
-}
-
-CLBuildOptions CLSaturatedArithmeticOperationKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
-{
-    const bool has_float_out = is_data_type_float(output.data_type());
-    auto       build_options = generate_build_options_with_arithmetic_rules(input1, input2, output, name());
-    build_options.add_option((_policy == ConvertPolicy::WRAP || has_float_out) ? "-DWRAP" : "-DSATURATE");
-    return build_options;
-}
-std::string CLSaturatedArithmeticOperationKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
-{
-    auto config_id = generate_id_for_tuning_common(kernel_name, input1, output);
-    config_id += (_policy == ConvertPolicy::WRAP) ? "_wrap_" : "_saturate_";
-    config_id += lower_string(string_from_data_layout(input1.data_layout()));
-    return config_id;
-}
-
-std::string CLSaturatedArithmeticOperationKernel::name()
-{
-    return supported_sat_arithmetic_ops[_op];
-}
-
-/** Arithmetic operations*/
-
-void CLArithmeticOperationKernel::configure(ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
-{
-    configure(CLKernelLibrary::get().get_compile_context(), op, input1, input2, output, act_info);
-}
-
-void CLArithmeticOperationKernel::configure(const CLCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output,
-                                            const ActivationLayerInfo &act_info)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_ERROR_THROW_ON(CLArithmeticOperationKernel::validate(op, input1, input2, output, act_info));
-    auto padding_info = get_padding_info({ input1, input2, output });
-
-    _op       = op;
-    _act_info = act_info;
-    configure_common(compile_context, input1, input2, output);
-    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
-}
-
-Status CLArithmeticOperationKernel::validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
-    if(op == ArithmeticOperation::DIV || op == ArithmeticOperation::POWER)
-    {
-        // Division and Power operators don't support integer arithmetic
-        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_float_only_supported_rules(*input1, *input2, *output));
-        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*input1->clone(), *input2->clone(), *output->clone()).first);
-    }
-    else
-    {
-        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
-        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
-    }
-    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled() && !is_data_type_float(output->data_type()));
-
-    return Status{};
-}
-std::pair<Status, Window> CLArithmeticOperationKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
-{
-    if(_op == ArithmeticOperation::DIV || _op == ArithmeticOperation::POWER)
-    {
-        // Division and Power operators don't support integer arithmetic
-        return validate_and_configure_window_for_division(input1, input2, output);
-    }
-    else
-    {
-        return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
-    }
-}
-
-CLBuildOptions CLArithmeticOperationKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
-{
-    return generate_build_options_with_arithmetic_rules(input1, input2, output, name());
-}
-std::string CLArithmeticOperationKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
-{
-    return generate_id_for_tuning_common(kernel_name, input1, output);
-}
-
-std::string CLArithmeticOperationKernel::name()
-{
-    return supported_arithmetic_ops[_op];
-}
-} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLElementwiseOperationKernel.h b/src/core/CL/kernels/CLElementwiseOperationKernel.h
deleted file mode 100644
index dd04a03..0000000
--- a/src/core/CL/kernels/CLElementwiseOperationKernel.h
+++ /dev/null
@@ -1,256 +0,0 @@
-/*
- * Copyright (c) 2018-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLELEMENTWISEOPERATIONKERNEL_H
-#define ARM_COMPUTE_CLELEMENTWISEOPERATIONKERNEL_H
-
-#include "arm_compute/core/Types.h"
-#include "src/core/CL/ICLKernel.h"
-#include "src/core/KernelTypes.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** Interface for an element-wise operation kernel
- *
- * Element-wise operation is computed by:
- * @f[ output(x,y) = OP(input1(x,y), input2(x,y))@f]
- *
- */
-class CLElementwiseOperationKernel : public ICLKernel
-{
-public:
-    /** Default constructor */
-    CLElementwiseOperationKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLElementwiseOperationKernel(const CLElementwiseOperationKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLElementwiseOperationKernel &operator=(const CLElementwiseOperationKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLElementwiseOperationKernel(CLElementwiseOperationKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLElementwiseOperationKernel &operator=(CLElementwiseOperationKernel &&) = default;
-    /** Default destructor */
-    ~CLElementwiseOperationKernel() = default;
-
-    // Inherited methods overridden:
-    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
-
-protected:
-    /** The name of the operation */
-    virtual std::string name() = 0;
-
-    /** Initialise the kernel's output.
-     *
-     * @param[in] input1 First tensor input info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/F16/U32/S32/F32.
-     * @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output Output tensor info. Data types supported: Same as @p input1.
-     *
-     * @return a pair of Status and Window
-     */
-    virtual std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) = 0;
-
-    /** Generate the build options for the specific kernel
-     *
-     * @reutrn a CLBuildOptions struct
-     */
-    virtual CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) = 0;
-
-    /** Generate the identifier for tuning
-     *
-     * @reutrn a string
-     */
-    virtual std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) = 0;
-
-    /** Commmon configure function for element-wise operators with no additional options (e.g., Div, Min, Max, SquaredDiff)
-     *
-     */
-    void configure_common(ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
-    /** Commmon configure function for element-wise operators with no additional options (e.g., Div, Min, Max, SquaredDiff)
-     *
-     */
-    void configure_common(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
-
-    ActivationLayerInfo _act_info;
-
-private:
-    const ITensorInfo *_input1; /**< Source tensor info 1 */
-    const ITensorInfo *_input2; /**< Source tensor info 2 */
-    ITensorInfo       *_output; /**< Destination tensor info */
-};
-
-class CLLogicalBinaryKernel : public CLElementwiseOperationKernel
-{
-public:
-    /** Default constructor */
-    CLLogicalBinaryKernel() = default;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLLogicalBinaryKernel(const CLLogicalBinaryKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLLogicalBinaryKernel &operator=(const CLLogicalBinaryKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLLogicalBinaryKernel(CLLogicalBinaryKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLLogicalBinaryKernel &operator=(CLLogicalBinaryKernel &&) = default;
-    /** Default destructor */
-    ~CLLogicalBinaryKernel() = default;
-    /** Function to configure kernel
-     *
-     * @param[in] compile_context The compile context to be used.
-     * @param[in] op              Logical binary operation to be executed.
-     * @param[in] input1          First tensor input info. Data types supported: U8.
-     * @param[in] input2          Second tensor input info. Data types supported: same as @p input1.
-     * @param[in] output          Output tensor info. Data types supported: same as @p input1.
-     */
-    void configure(const CLCompileContext &compile_context, kernels::LogicalOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output);
-    /** Static function to check if the given configuration is valid for this kernel
-     *
-     * @param[in] op     Logical binary operation to be executed.
-     * @param[in] input1 First tensor input info. Data types supported: U8.
-     * @param[in] input2 Second tensor input info. Data types supported: same as @p input1.
-     * @param[in] output Output tensor info. Data types supported: same as @p input1.
-     */
-    static Status validate(kernels::LogicalOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
-
-private:
-    // Inherited methods overridden:
-    std::string name() override;
-    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) override;
-    CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
-    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) override;
-
-    kernels::LogicalOperation _op{ kernels::LogicalOperation::Unknown };
-};
-
-/** Addition operation */
-class CLSaturatedArithmeticOperationKernel : public CLElementwiseOperationKernel
-{
-public:
-    CLSaturatedArithmeticOperationKernel()
-        : CLElementwiseOperationKernel(), _policy(), _op()
-    {
-    }
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel
-     *
-     * @param[in] op       Arithmetic operation to be executed.
-     * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2   Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output   Output tensor info. Data types supported: Same as @p input1.
-     * @param[in] policy   Policy to use to handle overflow.
-     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
-     */
-    void configure(ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel
-     *
-     * @param[in] compile_context The compile context to be used.
-     * @param[in] op              Arithmetic operation to be executed.
-     * @param[in] input1          First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2          Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output          Output tensor info. Data types supported: Same as @p input1.
-     * @param[in] policy          Policy to use to handle overflow.
-     * @param[in] act_info        (Optional) Activation layer information in case of a fused activation.
-     */
-    void configure(const CLCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy,
-                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLSaturatedArithmeticOperationKernel
-     *
-     * @param[in] op       Arithmetic operation to be executed.
-     * @param[in] input1   First tensor input info info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2   Second tensor input info info. Data types supported: Same as @p input1.
-     * @param[in] output   Output tensor info info. Data types supported: Same as @p input1.
-     * @param[in] policy   Policy to use to handle overflow.
-     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
-     *
-     * @return a Status
-     */
-    static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy,
-                           const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-protected:
-    // Inherited methods overridden:
-    std::string name() override;
-    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) override;
-    CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
-    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) override;
-
-private:
-    ConvertPolicy       _policy;
-    ArithmeticOperation _op;
-};
-
-class CLArithmeticOperationKernel : public CLElementwiseOperationKernel
-{
-public:
-    CLArithmeticOperationKernel()
-        : CLElementwiseOperationKernel(), _op()
-    {
-    }
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel
-     *
-     * @param[in] op       Arithmetic operation to be executed.
-     * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2   Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output   Output tensor info. Data types supported: Same as @p input1.
-     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
-     */
-    void configure(ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel
-     *
-     * @param[in] compile_context The compile context to be used.
-     * @param[in] op              Arithmetic operation to be executed.
-     * @param[in] input1          First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2          Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output          Output tensor info. Data types supported: Same as @p input1.
-     * @param[in] act_info        (Optional) Activation layer information in case of a fused activation.
-     */
-    void configure(const CLCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output,
-                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLArithmeticOperationKernel
-     *
-     * @param[in] op       Arithmetic operation to be executed.
-     * @param[in] input1   First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
-     * @param[in] input2   Second tensor input info. Data types supported: Same as @p input1.
-     * @param[in] output   Output tensor info. Data types supported: Same as @p input1.
-     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
-     *
-     * @return a Status
-     */
-    static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
-
-protected:
-    // Inherited methods overridden:
-    std::string name() override;
-    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) override;
-    CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
-    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) override;
-
-private:
-    ArithmeticOperation _op;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_CLELEMENTWISEOPERATIONKERNEL_H */
diff --git a/src/core/KernelTypes.h b/src/core/KernelTypes.h
index 12e6bc9..a32f5db 100644
--- a/src/core/KernelTypes.h
+++ b/src/core/KernelTypes.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,8 +26,6 @@
 
 namespace arm_compute
 {
-namespace kernels
-{
 /** List of supported logical operations */
 enum class LogicalOperation
 {
@@ -36,6 +34,5 @@
     Or,      /**< Logical Or || */
     Not,     /**< Logical Not ! */
 };
-} // namespace kernels
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_KERNEL_TYPES_H */
diff --git a/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp b/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp
new file mode 100644
index 0000000..7d204b1
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClElementwiseKernel.cpp
@@ -0,0 +1,505 @@
+/*
+ * Copyright (c) 2018-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "src/core/CL/CLValidate.h"
+#include "src/core/common/Validate.h"
+#include "src/core/helpers/AutoConfiguration.h"
+#include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
+#include "support/StringSupport.h"
+#include <map>
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+namespace
+{
+constexpr unsigned int vector_size_byte_opencl = 16;
+
+std::map<ArithmeticOperation, std::string> supported_arithmetic_ops =
+{
+    { ArithmeticOperation::ADD, "ADD" },
+    { ArithmeticOperation::SUB, "SUB" },
+    { ArithmeticOperation::DIV, "DIV" },
+    { ArithmeticOperation::SQUARED_DIFF, "SQUARED_DIFF" },
+    { ArithmeticOperation::MIN, "MIN" },
+    { ArithmeticOperation::MAX, "MAX" },
+    { ArithmeticOperation::POWER, "POWER" },
+    { ArithmeticOperation::PRELU, "PRELU" },
+};
+
+std::map<ArithmeticOperation, std::string> supported_sat_arithmetic_ops =
+{
+    { ArithmeticOperation::ADD, "ADD" },
+    { ArithmeticOperation::SUB, "SUB" },
+};
+
+std::string generate_id_for_tuning_common(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
+{
+    std::string config_id;
+    // Set config_id for enabling LWS tuning
+    config_id = kernel_name;
+    config_id += "_";
+    config_id += lower_string(string_from_data_type(src1.data_type()));
+    config_id += "_";
+    config_id += support::cpp11::to_string(dst.dimension(0));
+    config_id += "_";
+    config_id += support::cpp11::to_string(dst.dimension(1));
+    return config_id;
+}
+
+Status validate_arguments_with_float_only_supported_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(&src1, &src2, &dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&src1);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&src1, 1, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&src1, &src2);
+
+    const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
+
+    // Validate in case of configured dst
+    if(dst.total_size() > 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&dst, 1, DataType::F16, DataType::F32);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&src1, &dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst.tensor_shape(), 0),
+                                        "Wrong shape for dst");
+    }
+
+    return Status{};
+}
+
+Status validate_arguments_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&src1);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&src1, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
+                                                         DataType::S32, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&src2);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&src2, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
+                                                         DataType::S32, DataType::F32);
+
+    const bool is_quantized = is_data_type_quantized(src1.data_type()) || is_data_type_quantized(src2.data_type());
+    if(is_quantized)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&src1, &src2);
+
+        if(is_data_type_quantized_symmetric(src1.data_type()))
+        {
+            const int32_t in1_offset = src1.quantization_info().uniform().offset;
+            const int32_t in2_offset = src2.quantization_info().uniform().offset;
+            ARM_COMPUTE_RETURN_ERROR_ON_MSG(in1_offset != 0, "For quantized symmetric, offset must be zero");
+            ARM_COMPUTE_RETURN_ERROR_ON_MSG(in2_offset != 0, "For quantized symmetric, offset must be zero");
+        }
+    }
+
+    const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
+
+    // Validate in case of configured dst
+    if(dst.total_size() > 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&dst, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+                                                             DataType::S16, DataType::QSYMM16, DataType::F16,
+                                                             DataType::S32, DataType::F32);
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG((dst.data_type() == DataType::U8) && ((src1.data_type() != DataType::U8) || (src2.data_type() != DataType::U8)),
+                                        "dst can only be U8 if both inputs are U8");
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst.tensor_shape(), 0),
+                                        "Wrong shape for dst");
+
+        if(is_quantized)
+        {
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&src1, &dst);
+
+            if(is_data_type_quantized_symmetric(dst.data_type()))
+            {
+                const int32_t offset = dst.quantization_info().uniform().offset;
+                ARM_COMPUTE_RETURN_ERROR_ON_MSG(offset != 0, "For quantized symmetric, offset must be zero");
+            }
+        }
+    }
+    return Status{};
+}
+
+CLBuildOptions generate_build_options_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst, const std::string &operation_string)
+{
+    CLBuildOptions build_opts;
+
+    const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
+
+    build_opts.add_option("-DDATA_TYPE_IN1=" + get_cl_type_from_data_type(src1.data_type()));
+    build_opts.add_option("-DDATA_TYPE_IN2=" + get_cl_type_from_data_type(src2.data_type()));
+    build_opts.add_option("-DDATA_TYPE_OUT=" + get_cl_type_from_data_type(dst.data_type()));
+    build_opts.add_option("-DVEC_SIZE_IN1=" + support::cpp11::to_string(src1.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
+    build_opts.add_option("-DVEC_SIZE_IN2=" + support::cpp11::to_string(src2.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
+    build_opts.add_option("-DVEC_SIZE_OUT=" + support::cpp11::to_string(num_elems_processed_per_iteration));
+    build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(dst.dimension(0) % num_elems_processed_per_iteration));
+    build_opts.add_option("-DOP=" + operation_string);
+    if(is_data_type_quantized(src1.data_type()))
+    {
+        const UniformQuantizationInfo iq1info = src1.quantization_info().uniform();
+        const UniformQuantizationInfo iq2info = src2.quantization_info().uniform();
+        const UniformQuantizationInfo oqinfo  = dst.quantization_info().uniform();
+
+        build_opts.add_option("-DOFFSET_IN1=" + support::cpp11::to_string(iq1info.offset));
+        build_opts.add_option("-DOFFSET_IN2=" + support::cpp11::to_string(iq2info.offset));
+        build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(oqinfo.offset));
+        build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1info.scale));
+        build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2info.scale));
+        build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oqinfo.scale));
+    }
+    return build_opts;
+}
+
+std::pair<Status, Window> configure_window_arithmetic_common(ITensorInfo &dst)
+{
+    const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
+    Window             win                               = calculate_max_window(dst, Steps(num_elems_processed_per_iteration));
+    return std::make_pair(Status{}, win);
+}
+
+std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
+{
+    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
+    const TensorShape &out_shape = broadcast_pair.first;
+
+    set_shape_if_empty(dst, out_shape);
+
+    if(src1.data_type() == DataType::S16 || src2.data_type() == DataType::S16)
+    {
+        set_format_if_unknown(dst, Format::S16);
+    }
+    else if(src1.data_type() == DataType::F16 || src2.data_type() == DataType::F16)
+    {
+        set_format_if_unknown(dst, Format::F16);
+    }
+    else if(src1.data_type() == DataType::F32 || src2.data_type() == DataType::F32)
+    {
+        set_format_if_unknown(dst, Format::F32);
+    }
+    else if(src1.data_type() == DataType::QASYMM8 || src2.data_type() == DataType::QASYMM8)
+    {
+        set_data_type_if_unknown(dst, DataType::QASYMM8);
+    }
+    else if(src1.data_type() == DataType::QASYMM8_SIGNED || src2.data_type() == DataType::QASYMM8_SIGNED)
+    {
+        set_data_type_if_unknown(dst, DataType::QASYMM8_SIGNED);
+    }
+    else if(src1.data_type() == DataType::QSYMM16 || src2.data_type() == DataType::QSYMM16)
+    {
+        set_data_type_if_unknown(dst, DataType::QSYMM16);
+    }
+
+    return configure_window_arithmetic_common(dst);
+}
+
+std::pair<Status, Window> validate_and_configure_window_for_logical_binary_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
+{
+    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
+    const TensorShape &out_shape = broadcast_pair.first;
+
+    set_shape_if_empty(dst, out_shape);
+    set_data_type_if_unknown(dst, DataType::U8);
+
+    // The arithmetic utility functions can be share
+    return configure_window_arithmetic_common(dst);
+}
+
+std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
+{
+    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
+    const TensorShape &out_shape = broadcast_pair.first;
+    auto_init_if_empty(dst, out_shape, 1, src1.data_type());
+    return configure_window_arithmetic_common(dst);
+}
+} // namespace
+
+void ClElementwiseKernel::configure_common(ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
+{
+    configure_common(CLKernelLibrary::get().get_compile_context(), src1, src2, dst);
+}
+
+void ClElementwiseKernel::configure_common(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
+{
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(*src1, *src2, *dst);
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+
+    _src1 = src1;
+    _src2 = src2;
+    _dst  = dst;
+
+    std::string kernel_name = "elementwise_operation_" + name();
+    if(is_data_type_quantized(src1->data_type()))
+    {
+        kernel_name += "_quantized";
+    }
+
+    // Set kernel build options
+    CLBuildOptions build_opts = generate_build_options(*src1, *src2, *dst);
+    if(_act_info.enabled())
+    {
+        build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(_act_info.activation())));
+        build_opts.add_option("-DA_VAL=" + float_to_string_with_full_precision(_act_info.a()));
+        build_opts.add_option("-DB_VAL=" + float_to_string_with_full_precision(_act_info.b()));
+    }
+
+    // Create kernel
+    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
+
+    ICLKernel::configure_internal(win_config.second);
+
+    _config_id = generate_id_for_tuning(kernel_name, *src1, *dst);
+}
+
+void ClElementwiseKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    const auto src_0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
+    const auto src_1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
+    auto       dst   = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
+    const TensorShape &in_shape1 = src_0->info()->tensor_shape();
+    const TensorShape &in_shape2 = src_1->info()->tensor_shape();
+    const TensorShape &out_shape = dst->info()->tensor_shape();
+
+    bool       can_collapse = true;
+    const bool is_vector    = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
+    if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
+    {
+        can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
+        for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
+        {
+            can_collapse = (in_shape1[d] == in_shape2[d]);
+        }
+    }
+
+    bool   has_collapsed = false;
+    Window collapsed     = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
+
+    const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
+    const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
+
+    Window slice      = collapsed.first_slice_window_3D();
+    Window slice_src1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed);
+    Window slice_src2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed);
+    do
+    {
+        unsigned int idx = 0;
+        add_3D_tensor_argument(idx, src_0, slice_src1);
+        add_3D_tensor_argument(idx, src_1, slice_src2);
+        add_3D_tensor_argument(idx, dst, slice);
+
+        enqueue(queue, *this, slice, lws_hint());
+        ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src1));
+        ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src2));
+    }
+    while(collapsed.slide_window_slice_3D(slice));
+}
+
+/** Logical binary */
+
+void ClLogicalBinaryKernel::configure(const ClCompileContext &compile_context, LogicalOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(ClLogicalBinaryKernel::validate(op, src1, src2, dst));
+    _op = op;
+    configure_common(compile_context, src1, src2, dst);
+}
+
+Status ClLogicalBinaryKernel::validate(LogicalOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
+{
+    ARM_COMPUTE_UNUSED(op);
+    ARM_COMPUTE_ASSERT(op != LogicalOperation::Unknown && op != LogicalOperation::Not);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
+
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src1, 1, DataType::U8);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src1, src2);
+
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_logical_binary_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
+
+    return Status{};
+}
+
+std::string ClLogicalBinaryKernel::name()
+{
+    switch(_op)
+    {
+        case LogicalOperation::And:
+            return "AND";
+        case LogicalOperation::Or:
+            return "OR";
+        case LogicalOperation::Not:
+        /* fall through */
+        default:
+            ARM_COMPUTE_ASSERT(true);
+    }
+    return "";
+}
+
+std::pair<Status, Window> ClLogicalBinaryKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
+{
+    return validate_and_configure_window_for_logical_binary_operators(src1, src2, dst);
+}
+
+CLBuildOptions ClLogicalBinaryKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
+{
+    // The arithmetic utility functions can be share
+    return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
+}
+
+std::string ClLogicalBinaryKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
+{
+    return generate_id_for_tuning_common(kernel_name, src1, dst);
+}
+
+/** Arithmetic operations with saturation*/
+void ClSaturatedArithmeticKernel::configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output,
+                                            const ConvertPolicy       &policy,
+                                            const ActivationLayerInfo &act_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
+    ARM_COMPUTE_ERROR_THROW_ON(ClSaturatedArithmeticKernel::validate(op, input1, input2, output, policy, act_info));
+    auto padding_info = get_padding_info({ input1, input2, output });
+
+    _policy   = policy;
+    _op       = op;
+    _act_info = act_info;
+    configure_common(compile_context, input1, input2, output);
+    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+}
+
+Status ClSaturatedArithmeticKernel::validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy,
+                                             const ActivationLayerInfo &act_info)
+{
+    ARM_COMPUTE_UNUSED(op, policy);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled() && !is_data_type_float(output->data_type()));
+
+    return Status{};
+}
+
+std::pair<Status, Window> ClSaturatedArithmeticKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
+{
+    return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
+}
+
+CLBuildOptions ClSaturatedArithmeticKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
+{
+    const bool has_float_out = is_data_type_float(output.data_type());
+    auto       build_options = generate_build_options_with_arithmetic_rules(input1, input2, output, name());
+    build_options.add_option((_policy == ConvertPolicy::WRAP || has_float_out) ? "-DWRAP" : "-DSATURATE");
+    return build_options;
+}
+
+std::string ClSaturatedArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
+{
+    auto config_id = generate_id_for_tuning_common(kernel_name, input1, output);
+    config_id += (_policy == ConvertPolicy::WRAP) ? "_wrap_" : "_saturate_";
+    config_id += lower_string(string_from_data_layout(input1.data_layout()));
+    return config_id;
+}
+
+std::string ClSaturatedArithmeticKernel::name()
+{
+    return supported_sat_arithmetic_ops[_op];
+}
+
+/** Arithmetic operations*/
+void ClArithmeticKernel::configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst,
+                                   const ActivationLayerInfo &act_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(ClArithmeticKernel::validate(op, src1, src2, dst, act_info));
+    auto padding_info = get_padding_info({ src1, src2, dst });
+
+    _op       = op;
+    _act_info = act_info;
+    configure_common(compile_context, src1, src2, dst);
+    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
+}
+
+Status ClArithmeticKernel::validate(ArithmeticOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
+    if(op == ArithmeticOperation::DIV || op == ArithmeticOperation::POWER)
+    {
+        // Division and Power operators don't support integer arithmetic
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_float_only_supported_rules(*src1, *src2, *dst));
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*src1->clone(), *src2->clone(), *dst->clone()).first);
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
+    }
+    ARM_COMPUTE_RETURN_ERROR_ON(act_info.enabled() && !is_data_type_float(dst->data_type()));
+
+    return Status{};
+}
+std::pair<Status, Window> ClArithmeticKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
+{
+    if(_op == ArithmeticOperation::DIV || _op == ArithmeticOperation::POWER)
+    {
+        // Division and Power operators don't support integer arithmetic
+        return validate_and_configure_window_for_division(src1, src2, dst);
+    }
+    else
+    {
+        return validate_and_configure_window_for_arithmetic_operators(src1, src2, dst);
+    }
+}
+
+CLBuildOptions ClArithmeticKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
+{
+    return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
+}
+std::string ClArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
+{
+    return generate_id_for_tuning_common(kernel_name, src1, dst);
+}
+
+std::string ClArithmeticKernel::name()
+{
+    return supported_arithmetic_ops[_op];
+}
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClElementwiseKernel.h b/src/core/gpu/cl/kernels/ClElementwiseKernel.h
new file mode 100644
index 0000000..4ed8ae7
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClElementwiseKernel.h
@@ -0,0 +1,219 @@
+/*
+ * Copyright (c) 2018-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H
+#define ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H
+
+#include "src/core/KernelTypes.h"
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for an element-wise operation kernel
+ *
+ * Element-wise operation is computed by:
+ * @f[ dst(x,y) = OP(src1(x,y), src2(x,y))@f]
+ *
+ */
+class ClElementwiseKernel : public IClKernel
+{
+public:
+    /** Default constructor */
+    ClElementwiseKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClElementwiseKernel);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override;
+
+protected:
+    /** The name of the operation */
+    virtual std::string name() = 0;
+
+    /** Configure kernel for a given list of arguments
+     *
+     * @param[in] src1 First source tensor info. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/F16/U32/S32/F32.
+     * @param[in] src2 Second source tensor info. Data types supported: same as @p src1.
+     * @param[in] dst  Destination tensor info. Data types supported: same as @p src1.
+     *
+     * @return a pair of Status and Window
+     */
+    virtual std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) = 0;
+
+    /** Generate the build options for the specific kernel
+     *
+     * @reutrn a CLBuildOptions struct
+     */
+    virtual CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) = 0;
+
+    /** Generate the identifier for tuning
+     *
+     * @reutrn a string
+     */
+    virtual std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) = 0;
+
+    /** Commmon configure function for element-wise operators with no additional options (e.g., Div, Min, Max, SquaredDiff)
+     *
+     */
+    void configure_common(ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst);
+    /** Commmon configure function for element-wise operators with no additional options (e.g., Div, Min, Max, SquaredDiff)
+     *
+     */
+    void configure_common(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst);
+
+    ActivationLayerInfo _act_info{};
+
+private:
+    const ITensorInfo *_src1{ nullptr }; /**< Source tensor info 1 */
+    const ITensorInfo *_src2{ nullptr }; /**< Source tensor info 2 */
+    ITensorInfo       *_dst{ nullptr };  /**< Destination tensor info */
+};
+
+class ClLogicalBinaryKernel : public ClElementwiseKernel
+{
+public:
+    /** Default constructor */
+    ClLogicalBinaryKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClLogicalBinaryKernel);
+    /** Function to configure kernel
+     *
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] op              Logical binary operation to be executed.
+     * @param[in] src1            First source tensor info. Data types supported: U8.
+     * @param[in] src2            Second source tensor info. Data types supported: same as @p src1.
+     * @param[in] dst             Destination tensor info. Data types supported: same as @p src1.
+     */
+    void configure(const ClCompileContext &compile_context, LogicalOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst);
+    /** Static function to check if the given configuration is valid for this kernel
+     *
+     * @param[in] op   Logical binary operation to be executed.
+     * @param[in] src1 First source tensor info. Data types supported: U8.
+     * @param[in] src2 Second source tensor info. Data types supported: same as @p src1.
+     * @param[in] dst  Destination tensor info. Data types supported: same as @p src1.
+     */
+    static Status validate(LogicalOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst);
+
+private:
+    // Inherited methods overridden:
+    std::string name() override;
+    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) override;
+    CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) override;
+    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) override;
+
+    LogicalOperation _op{ LogicalOperation::Unknown };
+};
+
+/** Addition operation */
+class ClSaturatedArithmeticKernel : public ClElementwiseKernel
+{
+public:
+    ClSaturatedArithmeticKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClSaturatedArithmeticKernel);
+    /** Static function to check if given info will lead to a valid configuration of @ref ClSaturatedArithmeticKernel
+     *
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] op              Arithmetic operation to be executed.
+     * @param[in] input1          First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
+     * @param[in] input2          Second tensor input info. Data types supported: Same as @p input1.
+     * @param[in] output          Output tensor info. Data types supported: Same as @p input1.
+     * @param[in] policy          Policy to use to handle overflow.
+     * @param[in] act_info        (Optional) Activation layer information in case of a fused activation.
+     */
+    void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy,
+                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClSaturatedArithmeticKernel
+     *
+     * @param[in] op       Arithmetic operation to be executed.
+     * @param[in] input1   First tensor input info info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
+     * @param[in] input2   Second tensor input info info. Data types supported: Same as @p input1.
+     * @param[in] output   Output tensor info info. Data types supported: Same as @p input1.
+     * @param[in] policy   Policy to use to handle overflow.
+     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
+     *
+     * @return a Status
+     */
+    static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy,
+                           const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+protected:
+    // Inherited methods overridden:
+    std::string name() override;
+    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output) override;
+    CLBuildOptions generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output) override;
+    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output) override;
+
+private:
+    ConvertPolicy       _policy{};
+    ArithmeticOperation _op{};
+};
+
+class ClArithmeticKernel : public ClElementwiseKernel
+{
+public:
+    ClArithmeticKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClArithmeticKernel);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClArithmeticKernel
+     *
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] op              Arithmetic operation to be executed.
+     * @param[in] src1            First source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
+     * @param[in] src2            Second source tensor info. Data types supported: same as @p src1.
+     * @param[in] dst             Destination tensor info. Data types supported: same as @p src1.
+     * @param[in] act_info        (Optional) Activation layer information in case of a fused activation.
+     */
+    void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst,
+                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClArithmeticKernel
+     *
+     * @param[in] op       Arithmetic operation to be executed.
+     * @param[in] src1     First source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/F16/S32/F32.
+     * @param[in] src2     Second source tensor info. Data types supported: same as @p src1.
+     * @param[in] dst      Destination tensor info. Data types supported: same as @p src1.
+     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
+     *
+     * @return a Status
+     */
+    static Status validate(ArithmeticOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+protected:
+    // Inherited methods overridden:
+    std::string name() override;
+    std::pair<Status, Window> validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst) override;
+    CLBuildOptions generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst) override;
+    std::string generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst) override;
+
+private:
+    ArithmeticOperation _op{};
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_ELEMENTWISE_KERNEL_H */
diff --git a/src/runtime/CL/functions/CLElementwiseOperations.cpp b/src/runtime/CL/functions/CLElementwiseOperations.cpp
index a72e957..638990e 100644
--- a/src/runtime/CL/functions/CLElementwiseOperations.cpp
+++ b/src/runtime/CL/functions/CLElementwiseOperations.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,7 +25,9 @@
 
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
+
+#include "src/runtime/gpu/cl/operators/ClAdd.h"
 
 #include <utility>
 
@@ -33,34 +35,13 @@
 {
 namespace experimental
 {
-CLArithmeticAddition::CLArithmeticAddition()
-{
-}
-
-void CLArithmeticAddition::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
-{
-    auto k = std::make_unique<CLSaturatedArithmeticOperationKernel>();
-    k->configure(compile_context, ArithmeticOperation::ADD, input1, input2, output, policy, act_info);
-    _kernel = std::move(k);
-}
-
-Status CLArithmeticAddition::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
-{
-    return CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, input1, input2, output, policy, act_info);
-}
-
-void CLArithmeticAddition::run(ITensorPack &tensors)
-{
-    ICLOperator::run(tensors);
-}
-
 CLArithmeticSubtraction::CLArithmeticSubtraction()
 {
 }
 void CLArithmeticSubtraction::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, ConvertPolicy policy,
                                         const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLSaturatedArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClSaturatedArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::SUB, input1, input2, output, policy, act_info);
     _kernel = std::move(k);
 }
@@ -68,7 +49,7 @@
 Status CLArithmeticSubtraction::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
     ARM_COMPUTE_UNUSED(policy);
-    return CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::SUB, input1, input2, output, policy, act_info);
+    return arm_compute::opencl::kernels::ClSaturatedArithmeticKernel::validate(ArithmeticOperation::SUB, input1, input2, output, policy, act_info);
 }
 
 void CLArithmeticSubtraction::run(ITensorPack &tensors)
@@ -82,14 +63,14 @@
 
 void CLArithmeticDivision::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::DIV, input1, input2, output, act_info);
     _kernel = std::move(k);
 }
 
 Status CLArithmeticDivision::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::DIV, input1, input2, output, act_info);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::DIV, input1, input2, output, act_info);
 }
 
 void CLArithmeticDivision::run(ITensorPack &tensors)
@@ -103,14 +84,14 @@
 
 void CLElementwiseMax::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::MAX, input1, input2, output, act_info);
     _kernel = std::move(k);
 }
 
 Status CLElementwiseMax::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::MAX, input1, input2, output, act_info);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::MAX, input1, input2, output, act_info);
 }
 
 void CLElementwiseMax::run(ITensorPack &tensors)
@@ -124,14 +105,14 @@
 
 void CLElementwiseMin::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::MIN, input1, input2, output, act_info);
     _kernel = std::move(k);
 }
 
 Status CLElementwiseMin::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::MIN, input1, input2, output, act_info);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::MIN, input1, input2, output, act_info);
 }
 
 void CLElementwiseMin::run(ITensorPack &tensors)
@@ -145,14 +126,14 @@
 
 void CLElementwiseSquaredDiff::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::SQUARED_DIFF, input1, input2, output, act_info);
     _kernel = std::move(k);
 }
 
 Status CLElementwiseSquaredDiff::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::SQUARED_DIFF, input1, input2, output, act_info);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::SQUARED_DIFF, input1, input2, output, act_info);
 }
 
 void CLElementwiseSquaredDiff::run(ITensorPack &tensors)
@@ -166,14 +147,14 @@
 
 void CLElementwisePower::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::POWER, input1, input2, output, act_info);
     _kernel = std::move(k);
 }
 
 Status CLElementwisePower::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::POWER, input1, input2, output, act_info);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::POWER, input1, input2, output, act_info);
 }
 
 void CLElementwisePower::run(ITensorPack &tensors)
@@ -181,13 +162,12 @@
     ICLOperator::run(tensors);
 }
 } // namespace experimental
-
 struct CLArithmeticAddition::Impl
 {
-    const ICLTensor                                    *src_0{ nullptr };
-    const ICLTensor                                    *src_1{ nullptr };
-    ICLTensor                                          *dst{ nullptr };
-    std::unique_ptr<experimental::CLArithmeticAddition> op{ nullptr };
+    const ICLTensor               *src_0{ nullptr };
+    const ICLTensor               *src_1{ nullptr };
+    ICLTensor                     *dst{ nullptr };
+    std::unique_ptr<opencl::ClAdd> op{ nullptr };
 };
 
 CLArithmeticAddition::CLArithmeticAddition()
@@ -209,13 +189,13 @@
     _impl->src_0 = input1;
     _impl->src_1 = input2;
     _impl->dst   = output;
-    _impl->op    = std::make_unique<experimental::CLArithmeticAddition>();
+    _impl->op    = std::make_unique<opencl::ClAdd>();
     _impl->op->configure(compile_context, input1->info(), input2->info(), output->info(), policy, act_info);
 }
 
 Status CLArithmeticAddition::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ConvertPolicy policy, const ActivationLayerInfo &act_info)
 {
-    return experimental::CLArithmeticAddition::validate(input1, input2, output, policy, act_info);
+    return opencl::ClAdd::validate(input1, input2, output, policy, act_info);
 }
 
 void CLArithmeticAddition::run()
diff --git a/src/runtime/CL/functions/CLLogicalAnd.cpp b/src/runtime/CL/functions/CLLogicalAnd.cpp
index f1c5365..98c98ab 100644
--- a/src/runtime/CL/functions/CLLogicalAnd.cpp
+++ b/src/runtime/CL/functions/CLLogicalAnd.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,7 +23,7 @@
  */
 #include "arm_compute/runtime/CL/functions/CLLogicalAnd.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
 
 #include <utility>
 
@@ -33,14 +33,14 @@
 {
 void CLLogicalAnd::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
 {
-    auto k = std::make_unique<CLLogicalBinaryKernel>();
-    k->configure(compile_context, kernels::LogicalOperation::And, input1, input2, output);
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClLogicalBinaryKernel>();
+    k->configure(compile_context, LogicalOperation::And, input1, input2, output);
     _kernel = std::move(k);
 }
 
 Status CLLogicalAnd::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
-    return CLLogicalBinaryKernel::validate(kernels::LogicalOperation::And, input1, input2, output);
+    return arm_compute::opencl::kernels::ClLogicalBinaryKernel::validate(LogicalOperation::And, input1, input2, output);
 }
 
 void CLLogicalAnd::run(ITensorPack &tensors)
diff --git a/src/runtime/CL/functions/CLLogicalOr.cpp b/src/runtime/CL/functions/CLLogicalOr.cpp
index 8c6087e..897963a 100644
--- a/src/runtime/CL/functions/CLLogicalOr.cpp
+++ b/src/runtime/CL/functions/CLLogicalOr.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,7 +23,7 @@
  */
 #include "arm_compute/runtime/CL/functions/CLLogicalOr.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
 
 #include <utility>
 
@@ -33,14 +33,14 @@
 {
 void CLLogicalOr::configure(const CLCompileContext &compile_context, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output)
 {
-    auto k = std::make_unique<CLLogicalBinaryKernel>();
-    k->configure(compile_context, kernels::LogicalOperation::Or, input1, input2, output);
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClLogicalBinaryKernel>();
+    k->configure(compile_context, LogicalOperation::Or, input1, input2, output);
     _kernel = std::move(k);
 }
 
 Status CLLogicalOr::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
-    return CLLogicalBinaryKernel::validate(kernels::LogicalOperation::Or, input1, input2, output);
+    return arm_compute::opencl::kernels::ClLogicalBinaryKernel::validate(LogicalOperation::Or, input1, input2, output);
 }
 
 void CLLogicalOr::run(ITensorPack &tensors)
diff --git a/src/runtime/CL/functions/CLPReluLayer.cpp b/src/runtime/CL/functions/CLPReluLayer.cpp
index 876b5de..74286d4 100644
--- a/src/runtime/CL/functions/CLPReluLayer.cpp
+++ b/src/runtime/CL/functions/CLPReluLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,7 +21,7 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLElementwiseOperationKernel.h"
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
 
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
@@ -37,14 +37,14 @@
 
 void CLPReluLayer::configure(const CLCompileContext &compile_context, ITensorInfo *input, ITensorInfo *alpha, ITensorInfo *output)
 {
-    auto k = std::make_unique<CLArithmeticOperationKernel>();
+    auto k = std::make_unique<arm_compute::opencl::kernels::ClArithmeticKernel>();
     k->configure(compile_context, ArithmeticOperation::PRELU, input, alpha, output);
     _kernel = std::move(k);
 }
 
 Status CLPReluLayer::validate(const ITensorInfo *input, const ITensorInfo *alpha, const ITensorInfo *output)
 {
-    return CLArithmeticOperationKernel::validate(ArithmeticOperation::PRELU, input, alpha, output);
+    return arm_compute::opencl::kernels::ClArithmeticKernel::validate(ArithmeticOperation::PRELU, input, alpha, output);
 }
 
 void CLPReluLayer::run(ITensorPack &tensors)
diff --git a/src/runtime/NEON/functions/NELogical.cpp b/src/runtime/NEON/functions/NELogical.cpp
index 190998b..674ba40 100644
--- a/src/runtime/NEON/functions/NELogical.cpp
+++ b/src/runtime/NEON/functions/NELogical.cpp
@@ -50,7 +50,7 @@
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
 
     _impl->kernel = std::make_unique<kernels::NELogicalKernel>();
-    _impl->kernel->configure(input1->info(), input2->info(), output->info(), kernels::LogicalOperation::And);
+    _impl->kernel->configure(input1->info(), input2->info(), output->info(), LogicalOperation::And);
 
     _impl->pack = ITensorPack();
     _impl->pack.add_tensor(TensorType::ACL_SRC_0, input1);
@@ -60,7 +60,7 @@
 
 Status NELogicalAnd::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
-    return kernels::NELogicalKernel::validate(input1, input2, output, kernels::LogicalOperation::And);
+    return kernels::NELogicalKernel::validate(input1, input2, output, LogicalOperation::And);
 }
 
 void NELogicalAnd::run()
@@ -83,7 +83,7 @@
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
 
     _impl->kernel = std::make_unique<kernels::NELogicalKernel>();
-    _impl->kernel->configure(input1->info(), input2->info(), output->info(), kernels::LogicalOperation::Or);
+    _impl->kernel->configure(input1->info(), input2->info(), output->info(), LogicalOperation::Or);
 
     _impl->pack = ITensorPack();
     _impl->pack.add_tensor(TensorType::ACL_SRC_0, input1);
@@ -93,7 +93,7 @@
 
 Status NELogicalOr::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
-    return kernels::NELogicalKernel::validate(input1, input2, output, kernels::LogicalOperation::Or);
+    return kernels::NELogicalKernel::validate(input1, input2, output, LogicalOperation::Or);
 }
 
 void NELogicalOr::run()
@@ -116,7 +116,7 @@
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     _impl->kernel = std::make_unique<kernels::NELogicalKernel>();
-    _impl->kernel->configure(input->info(), nullptr, output->info(), kernels::LogicalOperation::Not);
+    _impl->kernel->configure(input->info(), nullptr, output->info(), LogicalOperation::Not);
 
     _impl->pack = ITensorPack();
     _impl->pack.add_tensor(TensorType::ACL_SRC_0, input);
@@ -125,7 +125,7 @@
 
 Status NELogicalNot::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
-    return kernels::NELogicalKernel::validate(input, nullptr, output, kernels::LogicalOperation::Not);
+    return kernels::NELogicalKernel::validate(input, nullptr, output, LogicalOperation::Not);
 }
 
 void NELogicalNot::run()
diff --git a/src/runtime/gpu/cl/operators/ClAdd.cpp b/src/runtime/gpu/cl/operators/ClAdd.cpp
new file mode 100644
index 0000000..01f550f
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClAdd.cpp
@@ -0,0 +1,47 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClAdd.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClElementwiseKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClAdd::configure(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst,
+                      ConvertPolicy policy, const ActivationLayerInfo &act_info)
+{
+    auto k = std::make_unique<kernels::ClSaturatedArithmeticKernel>();
+    k->configure(compile_context, ArithmeticOperation::ADD, src1, src2, dst, policy, act_info);
+    _kernel = std::move(k);
+}
+
+Status ClAdd::validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst,
+                       ConvertPolicy policy, const ActivationLayerInfo &act_info)
+{
+    return kernels::ClSaturatedArithmeticKernel::validate(ArithmeticOperation::ADD, src1, src2, dst, policy, act_info);
+}
+} // namespace opencl
+} // namespace arm_compute
diff --git a/src/runtime/gpu/cl/operators/ClAdd.h b/src/runtime/gpu/cl/operators/ClAdd.h
new file mode 100644
index 0000000..2854c16
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClAdd.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_ADD_H
+#define ARM_COMPUTE_CL_ADD_H
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run arithmetic addition
+ *
+ * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+ * @note The function performs an arithmetic addition between two tensors.
+ */
+class ClAdd : public IClOperator
+{
+public:
+    /** Default Constructor */
+    ClAdd() = default;
+    /** Configure function for a given list of arguments.
+     *
+     * Valid configurations (src1,src2) -> dst :
+     *
+     *   - (U8,U8)           -> U8
+     *   - (U8,U8)           -> S16
+     *   - (S16,U8)          -> S16
+     *   - (U8,S16)          -> S16
+     *   - (S16,S16)         -> S16
+     *   - (S32,S32)         -> S32
+     *   - (F16,F16)         -> F16
+     *   - (F32,F32)         -> F32
+     *   - (QASYMM8,QASYMM8) -> QASYMM8
+     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
+     *   - (QSYMM16,QSYMM16) -> QSYMM16
+     *
+     * @param[in]      compile_context The compile context to be used.
+     * @param[in, out] src1            First source tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     *                                 The source tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[in, out] src2            Second source tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     *                                 The source tensor is [in, out] because its TensorInfo might be modified inside the kernel in case of broadcasting of dimension 0.
+     * @param[out]     dst             Destination tensor. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     * @param[in]      policy          Policy to use to handle overflow.
+     * @param[in]      act_info        (Optional) Activation layer information in case of a fused activation.
+     */
+    void configure(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, ConvertPolicy policy,
+                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
+    /** Static function to check if given info will lead to a valid configuration of @ref ClAdd
+     *
+     * Valid configurations (src1,src2) -> dst :
+     *
+     *   - (U8,U8)           -> U8
+     *   - (U8,U8)           -> S16
+     *   - (S16,U8)          -> S16
+     *   - (U8,S16)          -> S16
+     *   - (S16,S16)         -> S16
+     *   - (S32,S32)         -> S32
+     *   - (F16,F16)         -> F16
+     *   - (F32,F32)         -> F32
+     *   - (QASYMM8,QASYMM8) -> QASYMM8
+     *   - (QASYMM8_SIGNED,QASYMM8_SIGNED) -> QASYMM8_SIGNED
+     *   - (QSYMM16,QSYMM16) -> QSYMM16
+     *
+     * @param[in] src1     First source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     * @param[in] src2     Second source tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     * @param[in] dst      Destination tensor info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/QSYMM16/S32/F16/F32.
+     * @param[in] policy   Policy to use to handle overflow.
+     * @param[in] act_info (Optional) Activation layer information in case of a fused activation.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, ConvertPolicy policy,
+                           const ActivationLayerInfo &act_info = ActivationLayerInfo());
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_ADD_H */
diff --git a/tests/validation/reference/Logical.cpp b/tests/validation/reference/Logical.cpp
index 099abf6..aab57d9 100644
--- a/tests/validation/reference/Logical.cpp
+++ b/tests/validation/reference/Logical.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,18 +34,18 @@
 namespace reference
 {
 template <typename T>
-T logical_binary_op(arm_compute::kernels::LogicalOperation op, T src1, T src2)
+T logical_binary_op(arm_compute::LogicalOperation op, T src1, T src2)
 {
     switch(op)
     {
-        case arm_compute::kernels::LogicalOperation::And:
+        case arm_compute::LogicalOperation::And:
             return src1 && src2;
-        case arm_compute::kernels::LogicalOperation::Or:
+        case arm_compute::LogicalOperation::Or:
             return src1 || src2;
         // The following operators are either invalid or not binary operator
-        case arm_compute::kernels::LogicalOperation::Not:
+        case arm_compute::LogicalOperation::Not:
         // fall through
-        case arm_compute::kernels::LogicalOperation::Unknown:
+        case arm_compute::LogicalOperation::Unknown:
         // fall through
         default:
             ARM_COMPUTE_ASSERT(true);
@@ -57,7 +57,7 @@
 struct BroadcastUnroll
 {
     template <typename T>
-    static void unroll(arm_compute::kernels::LogicalOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
+    static void unroll(arm_compute::LogicalOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
                        Coordinates &id_src1, Coordinates &id_src2, Coordinates &id_dst)
     {
         const bool src1_is_broadcast = (src1.shape()[dim - 1] != dst.shape()[dim - 1]);
@@ -84,7 +84,7 @@
 struct BroadcastUnroll<0>
 {
     template <typename T>
-    static void unroll(arm_compute::kernels::LogicalOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
+    static void unroll(arm_compute::LogicalOperation op, const SimpleTensor<T> &src1, const SimpleTensor<T> &src2, SimpleTensor<T> &dst,
                        Coordinates &id_src1, Coordinates &id_src2, Coordinates &id_dst)
     {
         dst[coord2index(dst.shape(), id_dst)] = logical_binary_op(op, src1[coord2index(src1.shape(), id_src1)], src2[coord2index(src2.shape(), id_src2)]);
@@ -99,7 +99,7 @@
     Coordinates     id_dst{};
     SimpleTensor<T> dst{ TensorShape::broadcast_shape(src1.shape(), src2.shape()), src1.data_type() };
 
-    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(arm_compute::kernels::LogicalOperation::Or, src1, src2, dst, id_src1, id_src2, id_dst);
+    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(arm_compute::LogicalOperation::Or, src1, src2, dst, id_src1, id_src2, id_dst);
 
     return dst;
 }
@@ -112,7 +112,7 @@
     Coordinates     id_dst{};
     SimpleTensor<T> dst{ TensorShape::broadcast_shape(src1.shape(), src2.shape()), src1.data_type() };
 
-    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(arm_compute::kernels::LogicalOperation::And, src1, src2, dst, id_src1, id_src2, id_dst);
+    BroadcastUnroll<Coordinates::num_max_dimensions>::unroll(arm_compute::LogicalOperation::And, src1, src2, dst, id_src1, id_src2, id_dst);
 
     return dst;
 }